SAS Guide With

Examples

Copyright mh2007

Table of Contents
Section 1 Preliminaries
1.1 Getting Started with SAS
1.2 More on Data Management by Example
Section 2 Parametric Tests
2.1 Getting Acquainted with the T-test
2.2 ANOVA by Example
2.3 Linear Regression by Example
Section 3 Nonparametric Tests
3.1 Wilcoxon Rank Sum Test and Kruskal-Wallis
Test by Example
Section 4 Logistic Regression
4.1 Logistic Regression by Example

Section 1

Preliminaries

1.1 Getting Started With SAS

One of the first things to consider prior to creating a data set is to know what information
you would like conveyed. The biggest problem that researchers have is coming up with a
clear and concise question and sticking to it. Under the assumption that the data of
interest has already been obtained through a primary source (your own data collection
technique) or a secondary source (another investigators data or maybe data on the
internet), you now must be able to bring this data into a usable format. This is referred to
as creating a data set. The following steps are helpful:
Using SAS
First, name the data set you would like to create.
data one;

The following data set is given the name, one. The data set, one, is empty for the
moment until actual raw data can be brought in.
Before going any further, one important thing to remember when using SAS is to put a
semicolon after each statement. This can get tedious at times, but is something that
should be kept in mind before starting anything in SAS.
In order to bring in your data a variety of possibilities are available. One method is to
keypunch your data directly into SAS. Before you start you must use the input command
as in the following line:
input Sub_name $ weight age;

The input command lets SAS know not only the variable name for each of the columns of
your data but also the type of data you will be bringing in. The two types of data that we
will be concerned with is character and numeric. From the input line, it is easily noticed
that Sub_name is your first column variable, and will refer to the subjects name. Since
names are character data, we need to let SAS know this. This is accomplished with the
use of the ($). Now SAS knows it will be reading in character or alpha numeric data for
the variable Sub_name. The other variables: height, weight, and treat, will be considered
numeric. One thing to note is that SAS is capable of reading in format types other than
character and numeric data. It can read in dates for example, but we will leave that issue
for a later discussion. Once SAS knows what type of data you will be using, and before
the keypunching process can begin a datalines statement is required.
datalines;

The datalines statement lets SAS know where the raw data exists and will be read from.
After the datalines statement, you can begin the keypunching process using a space as a
delimiter, which allows SAS to distinguish between the different variables you are
punching in data for. The following is an example of some observation lines:

JaySmith
MaryCarr
JoeJones
;

125 22
115 20
177 46

Notice that the data after being keypunched in was followed by a semicolon. This
indicates that there will be no more records entered and that SAS has reached the end of
the raw data set. Once these steps have been completed procedure statements called proc
statements can be used to view, manipulate, or analyze the data. Keep in mind that a
run statement usually follows all proc statements. You can use the following proc
statements to view the contents of what SAS has read into data set one:
proc print;
run;

If you are using the windows version of SAS, just point to the running man on the toolbar
and click. This is an execute command that runs the print procedure.
One thing to be aware of is that SAS by default will read only up to eight character
values. Some names longer than eight characters will get cut off at the eighth letter or
character. This problem is easily handled by specifying the number of characters that
should be allotted to accommodate the length of the longest observation being entered.
For example, assume that you would like to read in the following data set:
Jay Smith
Mary Carr
Joe Jones
Rachel Lonely
;

125
115
177
105

22
20
46
27

Now we have two problems that must be addressed. The first is that the subjects names
are more than eight characters in length. The other is that there are spaces between the
first and last names of each of the subjects. This is a problem because SAS uses a space
as a delimiter. When SAS encounters the space between the names, it assumes that the
next variable is being addressed which is not the case in our example. The following
input statement corrects our problem.
input Sub_name $ 1-13 weight age;

By specifying the number of spaces needed to accommodate the longest name, our
problem is solved. Specifying the number of spaces needed to read in data can be used
for both character and numeric data as follows:
input patient $ 1-13 weight 15-17 age 19-20;

Now try executing the following statements:

data one;
input patient $
datalines;
Jay Smith
125
Mary Carr
115
Joe Jones
177
Rachel Lonely 105
;
proc print;
run;

1-13 weight 15-17 age 19-20;

22
20
46
27

This is the output you should have obtained:

The SAS System
Obs
1
2
3
4

Sub_name
Jay Smith
Mary Carr
Joe Jones
Rachel Lonely

weight
125
115
177
105

age
22
20
46
27

Now lets explore some other ways of bringing in data.

Assume that the same information is saved as a text file, example.txt, in ASCII format
in the following location C:\Temp on your PC. This data set can be read using an
ASCII editor like notepad. The following is a copy of the contents stored in the file
example.txt:
Jay Smith
Mary Carr
Joe Jones
Rachel Lonely

125
115
177
105

22
20
46
27

Instead of copying and pasting this information from notepad to SAS, the data stored in
example.txt can be brought in using the following set of statements:
data one;
infile 'C:\Temp\example.txt';
input patient $ 1-13 weight 15-17 age 19-20;
proc print;
run;

The infile statement gives SAS the location of text file where the data of interest in
stored. Notice that the same input statement is used as if the data were keypunched
directly into SAS.
Now, assume that the same information is stored as an EXCEL file, example.xls, in the
following location C:\Temp on your PC. Although EXCEL is a good spreadsheet

program, it is not very accommodating if complex analysis of you data is required.

Therefore importing your data from EXCEL into SAS is often necessary, and can be done
easily with the following procedure statements:
proc import datafile = 'C:\Temp\example.xls' out = exdata1
replace;
proc print;
run;

The statement proc import allows the SAS user to import data from an EXCEL
spreadsheet into SAS. The datafile statement provides the reference location of the file.
In this case, the file example with the extension .xls to denote an EXCEL file is the
file we would like to import. The out statement is used to name the SAS data set that has
been created by the import procedure. Notice that print procedure has been utilized once
more in order to view the contents of the SAS data set exdata1.
Although we have mentioned character and numeric data types, we have yet to discuss
formatting numeric data, and how to work with variables that have dates as observations.
Looking at the following data set may help us understand the necessary steps involved
when working with dates as observations and formatting numeric data. The columns
in this data refers to the name of the subject, the date they were observed, their date
of birth, and their weight in pounds, which is to consist of two digits behind a
decimal point.
Mary
Joe
James

11/12/2002 06/05/78 12567

05/14/2001 07/08/67 15634
01/09/2002 02/28/64 16790

In this case the researcher would like to create a SAS data set using this information.
Later, the researcher would like to utilize the information consisting of the date of
observation and the date of birth to acquire the age of each subject. The researcher also
notices that the decimal point in the data for the weight of the subjects is missing.
Although these problems may seem complex to the novice SAS user, they can easily be
remedied with the use of informat statements. Informat statements allow the user to
specify the format of the variables at the input stage of the SAS code. This basically lets
SAS know what type of data it is reading in before it even reads it in.
The following is an input statement that contains informats and other symbols that we
have yet to discuss:
input sub $ 1-5 @8 obs_date MMDDYY10. @19 dob MMDDYY8. @28 weight d5.2;

The informat MMDDYY10. is commonly used when reading in dates. Notice that the
number of characters for the date of observation is ten, eight numeric values and the two
forward slashes. For the date of birth, there are eight characters, six numeric and the two
forward slashes. The variable weight uses the informat d5.2 which informs SAS to
read in five values with two behind a decimal point. Something else that should look
different is the @ followed by a number in front of the variable names. The @
indicates to SAS where to begin reading the variable observations that are being inputted.
For example @8, refers to character column 8, so SAS will begin reading obs_date
data at column eight.
At this point, we are ready to create a data set; lets call it testing, and it should look
something like the following:
data testing;
input sub $ 1-5
d5.2;
datalines;
Mary
11/12/2002
Joe
05/14/2001
James 01/09/2002
;
proc print;
run;

@8 obs_date MMDDYY10. @19 dob MMDDYY8. @28 weight

06/05/78 12567
07/08/67 15634
02/28/64 16790

After running this code, you should obtain the following output:
The SAS System
Obs

sub

obs_date

1
2
3

Mary
Joe
James

15656
15109
15349

dob

weight

6730
2745
1519

125.67
156.34
167.90

One thing that should be readily noticeable is that numeric values are given in place of
the dates. Those numeric values represent the number of days after January 1, 1960,
which is the year cutoff for SAS. Also note that the weight of the subjects is in the
desired format.
The calculation of the age for each subject is now relatively easy. By adding one line to
the previous SAS code, we can subtract the observation date from the date of birth and
divide by 365.25. I use 365.25 because it takes into account the fact that every four years
is a leap year. The results we desire should be provided with the following SAS code:
data testing;

input sub $ 1-5

d5.2;
age = (obs_date datalines;
Mary
11/12/2002
Joe
05/14/2001
James 01/09/2002
;
proc print;
run;

@8 obs_date MMDDYY10. @19

dob MMDDYY8. @28 weight

dob)/365.25;
06/05/78 12567
07/08/67 15634
02/28/64 16790

Notice that the line containing age is the only difference between this piece of SAS code
and the one previously written. The following is the output that should have been
obtained:

Obs

sub

1
2
3

Mary
Joe
James

The SAS System

obs_date
dob
15656
15109
15349

6730
2745
1519

weight

age

125.67
156.34
167.90

24.4381
33.8508
37.8645

1.2 More on Data Management by Example

The following tutorial allows you to go through some steps that will help you get better
acquainted with data management using SAS. First locate the file anthros.txt on the
diskette provided or on your hard drive. The file should contain the following lines of
text:
The following is a list of five individuals along with their
corresponding height in inches and weight in pounds.
Bruce Burks
Janie Johnson
Joe Jacobs
Jim Lewis
Linda Carol

66
62
72
70
60

128
135
157
186
119

There are quite a few things to consider before reading in this raw data. The first thing
that you should notice is that the first two lines give you information about your data set.
The third line is a blank line, and the data of interest starts on the fourth line.
Unfortunately, SAS cannot distinguish between text that is giving information about the
data and the data itself; however, this dilemma can easily be resolved in the following
manner:
data anthro;
infile 'C:\Temp\anthros.txt' firstobs = 4;
input Name $ 1-13 Height 15-16 Weight 18-20;

Here the data set that was created was named bmi. Next we used an INFILE statement to
bring in the raw data from either a diskette or our hard drive. Notice the option in the
INFILE statement firstobs. The firstobs option in the INFILE statement basically
alerts SAS to the fact that the data that should be read begins on the nth line. In our
example, the nth line is the fourth line, where n is restricted to the positive integers.
The firstobs option is extremely useful especially when you acquire data sets that have
extraneous information and want to jump directly into the analysis phase.
For the sake of completeness, lets take another example. The following text file contains
the exact same information as previously given the only difference is that its arranged
differently.
Bruce Burks
Janie Johnson
Joe Jacobs
Jim Lewis
Linda Carol

66
62
72
70
60

128
135
157
186
119

This is a list of five individuals along with their corresponding height

in inches and weight in pounds.

Here the first line contains the data we are interested in; however, if we try to infile this
data into SAS, the problem occurs in the last two lines. Without specifying where SAS
should stop reading the data, SAS will treat all the lines as if they were data that you
would like read. We can take care of this problem without modify the text file by using
the following command lines:
data anthro;
infile 'C:\Temp\anthros.txt' obs = 5;
input Name $ 1-13 Height 15-16 Weight 18-20;

In this case we are using the obs option, which alerts SAS to stop from reading any
more lines of data after the nth line has been read. In our example, we want SAS to stop
reading data after the fifth line has been read. You can easily view the contents of what
SAS has read in by using the proc print command followed by a run command. The
block of code should look like the following:
data anthro;
infile 'C:\Temp\anthros.txt' obs = 5;
input Name $ 1-13 Height 15-16 Weight 18-20;
proc print;
run;

In order to appreciate the significance of what has been learned so far, lets extend our
example into one that is a bit more complex. Lets assume that the same researcher has
assigned two employees into the field in order to gather more data on the anthropometrics
of the individuals eating at fast food restaurants.
The first employee returns with the following data set stored in the text file
anthros1.txt:
I went to the following places: BurgerKing, McDonalds, and Pizza Hut.
randomly chose five individuals and collected their name, height in
inches, and weight in pounds. These are my results:
Brian Long
Jim Branda
Brenda Green
Jo Ling
Daren Woods

72
69
59
62
68

208
200
157
112
156

The researcher, after opening the file, knows that the following SAS code is needed in
order to read in the data:
data one;
infile 'C:\Temp\anthros1.txt' firstobs = 5;
input Name $ 1-12 Height 14-15 Weight 17-19;

Now the second employee arrives and gives the researcher his data set named
anthros2.txt containing the following information:
Jim Jacobs
Wendall Lee
Pat Jones
Nel Henry
Jeremy Jacks

69
70
60
73
67

165
220
143
210
182

I went to Wendy's and Jack-in-the-Box.

random.

Five individuals were chosen at

Again, the researcher, after opening the file, knows that the following SAS code is
needed in order to read the data in:
data two;
infile 'C:\Temp\anthros2.txt' obs = 5;
input Name $ 1-12 Height 14-15 Weight 17-19;

Now that the researcher has these two data sets, she would like to bring them together to
form one big data set. This is called concatenation and SAS is capable of concatenating
data sets provided the variable names and content coincide among data sets. The
following code illustrates how this is done:
data one;
infile 'C:\Temp\anthros1.txt' firstobs = 5;
input Name $ 1-12 Height 14-15 Weight 17-19;
run;
data two;
infile 'C:\Temp\anthros2.txt' obs = 5;
input Name $ 1-12 Height 14-15 Weight 17-19;
run;
data three;
set one two;
proc print;
run;

The set command is critical here because it allows you to set data sets one and two into
a third data set called three. The results can be seen in the output after running proc
print:
The SAS System
Obs
1

Name
Brian Long

Height
72

Weight
208

2
3
4
5
6
7
8
9
10

Jim Branda
Brenda Green
Jo Ling
Daren Woods
Jim Jacobs
Wendall Lee
Pat Jones
Nel Henry
Jeremy Jacks

69
59
62
68
69
70
60
73
67

200
157
112
156
165
220
143
210
182

Finally, we will end this discussion with merging data sets. Lets assume that a third
employee hired by the researcher had accompanied the other two employees and recorded
the ages of the individuals who were interviewed. However, this information was not put
into data set three, the previous data set that was formed. First, lets look at the contents
of the text file named ages.txt submitted by the third employee:
Brian Long
Jim Branda
Brenda Green
Jo Ling
Daren Woods
Jim Jacobs
Wendall Lee
Pat Jones
Nel Henry
Jeremy Jacks

23
64
51
20
33
47
55
22
19
44

Since the goal is to merge data set three with data set ages, the merge can be
accomplished but a by variable must be specified. The by variable is usually
information on observations that is common to both data sets. In this case, the only
choice for a by variable would be the variable Name, since the names are given in
both sets of data. Before we proceed, it should be mentioned that the data sets must be
sorted by the by variable, in this case Name, prior to the merge. The following
command lines illustrate this procedure:
data one;
infile 'C:\Temp\anthros1.txt' firstobs = 5;
input Name $ 1-12 Height 14-15 Weight 17-19;
run;
data two;
infile 'C:\Temp\anthros2.txt' obs = 5;
input Name $ 1-12 Height 14-15 Weight 17-19;
run;
data three;
set one two;
proc sort;
by Name;
run;
data four;
infile 'C:\Temp\ages.txt';
input Name $ 1-12 Age 14-15;

proc sort;
by Name;
proc;
run;
data five;
merge three four;
by Name;
proc print;
run;

If you will notice, the data sets of interest are three and four since they contain the
information that we would like merged. The previous two data sets, data sets one and
two, were also included for completeness since they were utilized in the creation of data
set three. After running this procedure, the output obtained should appear as follows:
The SAS System
Obs
1
2
3
4
5
6
7
8
9
10

Name
Brenda Green
Brian Long
Daren Woods
Jeremy Jacks
Jim Branda
Jim Jacobs
Jo Ling
Nel Henry
Pat Jones
Wendall Lee

Height
59
72
68
67
69
69
62
73
60
70

Weight
157
208
156
182
200
165
112
210
143
220

Age
51
23
33
44
64
47
20
19
22
55

Notice that the last column contains the information on the ages of each of the individuals
observed.

Section 2

Parametric Tests

2.1

Getting Acquainted with the T-test

This tutorial will introduce you to t-tests, ANOVA (Analysis of Variance), and basic
linear regression. Lets begin with a t-test.
One common application of the t-test is to test whether the means of two different
variables, measured over the same set of observations, have a significant difference. This
is basically the same thing as saying does the mean of the difference between the two
variables of interest, over the same set of observations, differ significantly from zero.
This can best be illustrated via an example. Take for instance the data collected on
individuals heights and weights from different fast food restaurants. The following
information comes from the first employee:
Brian Long
Jim Branda
Brenda Green
Jo Ling
Daren Woods

72
69
59
62
68

208
200
157
112
156

And the next piece of information comes from the second employee:
Jim Jacobs
Wendall Lee
Pat Jones
Nel Henry
Jeremy Jacks

69
70
60
73
67

165
220
143
210
182

Lets assume that we have read the tutorial on data set management, or have some
familiarity with SAS already. We use our techniques to concatenate the two data sets and
obtain the following data set:
Brian Long
Jim Branda
Brenda Green
Jo Ling
Daren Woods
Jim Jacobs
Wendall Lee
Pat Jones
Nel Henry
Jeremy Jacks

72
69
59
62
68
69
70
60
73
67

208
200
157
112
156
165
220
143
210
182

Now we would like to distinguish between the two groups. One way to do this is by
adding another variable called Group. Group can be labeled 1 for the first set of five
observations and 2 for the second set.

Before we continue we should save the information we have as text in a temporary file in
our C-drive with the name onettest. The .txt is implied so when saving you shouldnt
have to add it on. Now, the following SAS statements can be used to create the data set
ttesting:
data ttesting;
infile 'C:\Temp\onettest.txt';
input Name $ 1-12 Height 14-15 Weight 17-19 Group 21-21;

Lets assume that the researcher would like to see if there is a significant difference in the
mean weight of individuals interviewed from each of the restaurants by his employees.
The following commands can be used to run a t-test:
proc ttest;
class Group;
var Weight;
run;

The class command contains the variable that we are using to distinguish between the
two groups of observations collected by each employee. Since we are interested in
comparing the weights of the two groups, we use the var command with the specified
variable of interest. In this case, the variable Weight as shown above. One more thing
that should be noted is that although we have the same number of individual observation
for each group this is not a paired t-test. You should run a pair t-test only when there is
some logical reasoning for pairing the variables of interest. One case when pairing is
required is when one observation is dependent on another observation in order to obtain
the correct information desired. For example, Assume you have a group of student and
that you give them a pre-test at the beginning of the semester and a post-test at the end.
Now, you are interested to see if the course increased their knowledge. The best way to
do this is to pair the post-test to the pre-test of each student, and examine the difference in
his or her scores. An example will be given later to demonstrate the details.
Now getting back to our example, we would like to see what kind of output SAS will
provide us with. Therefore after running the previous commands we acquire the
following results:
The SAS System
The TTEST Procedure
Statistics
Variable
Weight
Weight
Weight

Class

N
1
2

Diff (1-2)

5
5

Lower CL
Mean

Mean

Upper CL
Mean

Lower CL
Std Dev

Std Dev

Upper CL
Std Dev

Std Err

118.45
144.65
-69.05

166.6
184
-17.4

214.75
223.35
34.249

23.234
18.989
23.921

38.779
31.694
35.414

111.43
91.074
67.845

17.342
14.174
22.398

Variable
Weight
Weight

Method
Pooled
Satterthwaite

T-Tests
Variances
Equal
Unequal

DF
8
7.7

t Value
-0.78
-0.78

Pr > |t|
0.4596
0.4605

Equality of Variances
Variable
Weight

Method
Folded F

Num DF
4

Den DF
4

F Value
1.50

Pr > F
0.7054

One of the problems that most individuals complain about when using SAS is that it
gives too much output, or output that most individuals find extraneous. Although this is
true most of the time, the output displayed here is very useful. However, we are
interested in the Weight Diff, which is the difference in the means of the two weights.
We would like to know if the difference is significantly different from zero. Under the
null hypothesis, we are assuming no significant difference. Also, notice that SAS
provides the statistics for the assumption of equality of variance. In this case with a pvalue of 0.7054, we cannot reject the null hypothesis that the variances for the two groups
are equal. Now, observing the t-test with the pooled method, we obtain a p-value of
0.4596 and thus cannot reject the null hypothesis of no difference between the mean
weights of the two groups. One more thing that should be mentioned is that the
hypothesis test we conducted was a two-tailed test. Basically what this means is that we
began with the null hypothesis of the mean difference being equal to zero, and the
alternative hypothesis that the mean difference is not equal to zero. Since no
directionality is implied meaning that one group is not expected to have a higher mean
than the other, we refer to our test as a two-tailed test rather than a one-tailed test.
Now lets explore another scenario. The next data set was formed after the individuals
who were interviewed regarding their height and weight were exposed to a six-week diet
intervention. Their weights were recorded prior to the six-week period, and after they
completed the intervention. The following is a copy of the data set after this information
was collected:
Brian Long
Jim Branda
Brenda Green
Jo Ling
Daren Woods
Jim Jacobs
Wendall Lee
Pat Jones
Nel Henry
Jeremy Jacks

72
69
59
62
68
69
70
60
73
67

208
200
157
112
156
165
220
143
210
182

200
198
160
110
150
168
219
140
204
183

Now the researcher is interested to see if the intervention may have had an impact on the
weight of the individuals. The following SAS code can be used to run a paired t-test in
order to acquire the desired information:
data ttesting;
infile 'C:\Temp\pairttest.txt';

input Name $ 1-12 Height 14-15 Weight1 17-19 Weight2 21-23;

proc ttest;
paired Weight1*Weight2;
run;

The corresponding output looks like the following:

The SAS System
The TTEST Procedure
Statistics
Difference
Weight1 - Weight2

N
10

Lower CL
Mean
-0.607

Mean
2.1

Upper CL
Mean
4.8072

Lower CL
Std Dev
2.6031

Std Dev
3.7845

Upper CL
Std Dev
6.909

Std Err
1.1968

T-Tests
Difference
Weight1 - Weight2

DF
9

t Value
1.75

Pr > |t|
0.1132

Notice here that once again we cannot reject the null hypothesis, and must conclude that
the intervention had no significant impact on weight.

2.2 ANOVA by Example

Next, we look at example demonstrating the use of ANOVA. Lets assume that the
researcher has data on individuals from three different diet camps. All the researcher is
concerned with is seeing whether the mean weights of the individuals in each camp are
significantly different from one another. Since we are comparing three different means,
we must employ the use of ANOVA. The proceeding lines represent the SAS code and
data for this example.
data expanova;
input group weight;
datalines;
1 223
1 234
1 254
1 267
1 234
2 287
2 213
2 215
2 234
2 256
3 234
3 342
3 198
3 256
3 303
;
proc anova;
class group;
model weight = group;
means group;
run;

The following is the corresponding output to the previous lines of code.

The SAS System
The ANOVA Procedure
Dependent Variable: weight
Source
Model
Error
Corrected Total
R-Square
0.103219
Source
group

Sum of
DF
Squares
Mean Square
F Value
2
2071.60000
1035.80000
0.69
12
17998.40000
1499.86667
14
20070.00000
Coeff Var
Root MSE
weight Mean
15.49124
38.72811
250.0000
DF
Anova SS
Mean Square
F Value
2
2071.600000
1035.800000
0.69

Here we see that because the model is not significant (p-value = 0.5201), we must
conclude that the mean weights of the three groups are not significantly different.

Pr > F
0.5201

Pr > F
0.5201

2.3 Linear Regression by Example

Finally, lets end our discussion with basic linear regression. The researcher has
information on the distance individuals live from fast-food places and the number of
times they eat at those places per week. The researcher wants to see if there is a
relationship between these to variables in order to assess if location of residence is a good
predictor of the number of times individuals eat at fast-food places per week. The
following is a view of the raw data saved as reg.txt in a Temp file on the C-drive. The
first column represents the number of miles from the fast-food place (miles) and the
second column is the number of times per week the individual eats at the fast-food place
(eat_times). Next, you will see the SAS code used to access the data and run the linear
regression.
0.34
1.23
6.78
5.54
0.43
1.18
2.34
0.98
3.34
1.20

5
3
1
1
4
4
2
4
2
4

data fastfood;
infile 'C:\Temp\reg.txt';
input miles eat_times;
proc reg;
model eat_times = miles;
plot eat_times*miles;

run;
Finally, the SAS output is shown with a plot of the data.
The SAS System
The REG Procedure
Model: MODEL1
Dependent Variable: eat_times
Analysis of Variance
DF

Sum of
Squares

Mean
Square

1
8
9

15.07435
2.92565
18.00000

15.07435
0.36571

Root MSE
Dependent Mean
Coeff Var

0.60474
3.00000
20.15788

Source
Model
Error
Corrected Total

R-Square
Adj R-Sq

F Value

Pr > F

41.22

0.0002

0.8375
0.8171

Parameter Estimates
Variable
Intercept
miles

DF

Parameter
Estimate

Standard
Error

t Value

Pr > |t|

1
1

4.36223
-0.58315

0.28564
0.09083

15.27
-6.42

<.0001
0.0002

In this example the number of times individuals ate at the fast-food places decreased with
the number of miles they live from the fast-food place. The model is significant with a pvalue = 0.0002 and the relationship is strong since R-squared = 0.8375.

Section 3

NonParametric Tests

3.1 Wilcoxon Rank Sum Test and Kruskal-Wallis Test by Example

This is a tutorial on nonparametric tests. The biggest difference between a parametric
and nonparametric test is the fact that a parametric test assumes that the data under
investigation is coming from a normal distribution. The SAS software provides several
nonparametric tests such as the Wilcoxon rank-sum test, which is equivalent to the MannWhitney test for two samples, and the Kruskal-Wallis test when dealing with two or more
samples.
Consider the following example, Researcher Bob is interested in testing the difference
between the effectiveness of two allergy drugs out on the market. He would like to
administer drug A to a random sample of study subjects and then drug B to another
random sample who suffer from the same symptoms as those individuals taking drug A.
Researcher Bob would like to see if there is a difference between the two groups in the
time, in minutes, for subjects to feel relief from their allergy symptoms.
Table 1. The time, in minutes, until relief is felt after drug has been administered for
study groups A and B.
Subject
Drug A
Drug B
1
43
28
2
40
33
3
32
48
4
37
37
5
55
40
6
50
42
7
52
35
8
33
43
In order to do some analysis the first step would be to form a SAS data set as follows:
data drugtest;
input subject drug_group $ time;
datalines;
1 A 43
2 A 40
3 A 32
4 A 37
5 A 55
6 A 50
7 A 52
8 A 33
9 B 28
10 B 33
11 B 48
12 B 37
13 B 40
14 B 42
15 B 35
16 B 43

;
proc means median min max;
by drug_group;
var time;
proc npar1way wilcoxon;
class drug_group;
var time;
run;

In order to perform these nonparametric tests, we use the procedure npar1way and the
option wilcoxon to request that particular test. Note that wilcoxon is not the only
option available. Other options are median, mood, and exact, but for our purposes
we will not discuss these options at this time. In the case above, we have demonstrated
the Wilcoxon Rank Sum Test in order to test if a significant difference exists between the
drug groups in median time to relief from allergy symptoms. The following is the output
obtained from the SAS code:
------------------------------------------- drug_group=A
Analysis Variable : time
Median
Minimum
Maximum

41.5000000
32.0000000
55.0000000

------------------------------------------- drug_group=B
Analysis Variable : time
Median
Minimum
Maximum

38.5000000
28.0000000
48.0000000

The NPAR1WAY Procedure

Wilcoxon Scores (Rank Sums) for Variable time
Classified by Variable drug_group
drug_
Sum of
Expected
Std Dev
Mean
group
N
Scores
Under H0
Under H0
Score

A
8
77.0
68.0
9.493858
9.6250
B
8
59.0
68.0
9.493858
7.3750
Average scores were used for ties.
Wilcoxon Two-Sample Test
Statistic

77.0000

Normal Approximation
Z
One-Sided Pr > Z
Two-Sided Pr > |Z|

0.8953
0.1853
0.3706

t Approximation
One-Sided Pr > Z
Two-Sided Pr > |Z|

0.1924
0.3848

Z includes a continuity correction of 0.5.

Kruskal-Wallis Test
Chi-Square
0.8987

DF
Pr > Chi-Square

1
0.3431

The first thing that you should notice is that some descriptive statistics where given prior
to the test of median differences. This is mainly because procedure npar1way displays
the mean not the median for each drug group. Finally, we are interested in observing the
output given under the heading Normal Approximation. The probabilities given are those
associated with the Mann-Whitney test. Because of the way this hypothetical problem is
presented, a two-tailed test should be employed mainly because the directionality of the
outcome is not known (you dont know which drug works better). In any case, because
the probabilities are above 0.05, you cannot reject the null hypothesis and must conclude
that there is no difference between the median times to relief for both drug groups.

Section 4

Logistic Regression

4.1 Logistic Regression by Example

The following is a tutorial on logistic regression. There are several books that discuss
logistic regression in detail, but for our purposes well just skim the surface. Logistic
regression is a powerful tool; however, knowing when to use it is the most important
thing. Fundamentally, it is said that logistic regression can be used when you have an
outcome variable that is dichotomous. Although that is usually the case, it isnt the only
time. You can use logistic regression with polychotomous and ordinal data also.
Before we begin, lets discuss the linear regression model in order to gain a better
understanding of the differences between linear and logistic regression. The standard
assumptions of linear regression using ordinary least squares are as follows:

The model has the form: yi = b0 + b1x + ei

The expected value of ei = 0
The variance of ei is constant
The eis are independent and identically distributed as a Normal distribution

For logistic regression, these assumptions change in the following manner:

The expected value of ei = pi

The variance of ei = pi(1 - pi)
The eis are binomially distributed

It is critical that we look at these assumptions in order to get a better understanding of our
analysis. Now lets begin with a simple example in order to introduce the topic in better
detail. Assume that Researcher Smith has conducted a study involving suicide and
depression. Researcher Smith is interested in this relationship between suicide and
depression because he would like to show that individuals who are depressed are more
likely to attempt suicide. The following data was collected via surveys from a psychiatric
ward in smalltown. Note that 1 for depression indicates that the individual was
considered clinically depressed because they scored above a 5 on a depression criteria
scale. A 1 for suicide indicates that the individual attempted suicide. Data on 20
individuals is given below.
Subject Suicide DepressionCriteria Depression
1 1 4 0
2 0 2 0
3 1 4 0
4 0 3 0
5 0 6 1
6 0 8 1
7 1 4 0
8 0 2 0
9 0 2 0
10 0 3 0
11 0 2 0

12
13
14
15
16
17
18
19
20

1
0
0
1
1
1
1
1
1

4
1
6
7
9
7
8
9
8

0
0
1
1
1
1
1
1
1

Proceed to view your data with the following commands:

data suicide;
infile 'C:\Temp\suicide.txt' firstobs = 2;
input subject suicide depcritera depression;
proc print;
run;

One thing that can be done now is a plot of the data. At the moment we will not be
discussing graphics using SAS. Therefore we will continue the discussion in Excel just
to obtain a plot of the data and resume with SAS. For the moment, we will assume that
you have some experience with Excel and are capable of manipulating a spreadsheet and
making charts. Proceed by making a chart that plots Suicide vs Depression. Your plot
should look like the following:

1.2

Attempted Suicide

1
0.8
0.6
0.4
0.2
0
0

10

Depression Scale

At this point, it should be relatively easy to see that a straight line will not model this type
of data very well. However, an S-shaped curve or sigmoid curve would do the trick. For
that reason, we would employ logistic regression. In order to make our example easier

we will consider the variables suicide and depression which are both characterized as 1s
and 0s. With the following SAS code, we can run a logistic regression as follows:
data suicide;
infile 'C:\Temp\suicide.txt' firstobs = 2;
input subject suicide depcriteria depression;
proc logistic descending;
model suicide = depression;
run;

The only thing that is different in this block of code is the proc logistic which initiates
the logistic procedure. Notice the option descending after the procedure statement.
This tells SAS to use 1 as the outcome of interest for the dependent variable, suicide.
After the proc statement comes the statement containing the model format, you should
always put the dependent variable first, an equal sign, and finally the independent
variable(s).
The following output is obtained once these commands have been executed:
The LOGISTIC Procedure
Model Information
Data Set
Response Variable
Number of Response Levels
Number of Observations
Link Function
Optimization Technique

WORK.SUICIDE
suicide
2
20
Logit
Fisher's scoring

Response Profile
Ordered
Value

suicide

Total
Frequency

1
2

1
0

10
10

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.
Model Fit Statistics

Criterion
AIC
SC
-2 Log L

Intercept
Only

Intercept
and
Covariates

29.726
30.722
27.726

29.878
31.869
25.878

Testing Global Null Hypothesis: BETA=0

Test
Likelihood Ratio
Score
Wald

Chi-Square

DF

Pr > ChiSq

1.8480
1.8182
1.7574

1
1
1

0.1740
0.1775
0.1850

Analysis of Maximum Likelihood Estimates

Parameter
Intercept
depression

DF

Estimate

Standard
Error

Chi-Square

Pr > ChiSq

1
1

-0.5596
1.2526

0.6268
0.9449

0.7971
1.7574

0.3720
0.1850

Odds Ratio Estimates

Point
Estimate

Effect
depression

3.500

95% Wald
Confidence Limits
0.549

22.300

Association of Predicted Probabilities and Observed Responses

Percent Concordant
Percent Discordant
Percent Tied
Pairs

42.0
12.0
46.0
100

Somers' D
Gamma
Tau-a
c

0.300
0.556
0.158
0.650

Now that you have had a chance to view the output pay special attention to the section
that is in bold titled Odds Ratio Estimates. From this information we get an estimate for
the OR = 3.5. Basically this tells us that individuals who attempted suicide are 3.5 times
more likely to be depressed. However, you must pay close attention to the 95%
Confidence Interval of (0.549, 22.300) that accompanies the OR. Because the 95%
Confidence Interval contains 1.000, the estimate for the OR is not statistically significant.
Now lets take the following scenario. Assume that Researcher Smith believes that Race
is a key factor in suicide attempts, and he would like to explore his hypothesis by looking
at data on suicide and depression coming from Whites, Blacks, and Hispanics. We will
be using the text file suicide2.txt. You should have been given a diskette where this
data is stored. The researcher knows that is it common practice to use Whites as the
reference group. In this example, we would like to compare Blacks and Hispanics to
Whites, but in order to do so we need to create dummy variables. Table 1 refers to how
this is to be done.
Table1.
Whites (ref)
Blacks
Hispanics

dummy1
0
1
0

dummy2
0
0
1

At the moment, this may seem confusing; however, the following SAS code should give
you some insight when creating dummy variables:
data suicide2;
infile 'A:\suicide2.txt' firstobs = 2;
input Suicide Depression Race;
if Race = 2 then dummy1 = 1;

else dummy1 = 0;
if Race = 3 then dummy2 = 1;
else dummy2 = 0;
proc logistic descending;
model Suicide = dummy1 dummy2;
run;

After the program is run, the following output is obtained:

The LOGISTIC Procedure
Model Information
Data Set
Response Variable
Number of Response Levels
Number of Observations
Link Function
Optimization Technique

WORK.SUICIDE2
Suicide
2
105
Logit
Fisher's scoring

Response Profile
Ordered
Value

Suicide

Total
Frequency

1
2

1
0

42
63

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.
Model Fit Statistics

Criterion
AIC
SC
-2 Log L

Intercept
Only

Intercept
and
Covariates

143.332
145.986
141.332

142.199
150.160
136.199

Testing Global Null Hypothesis: BETA=0

Test

Chi-Square

DF

Pr > ChiSq

5.1339
5.0000
4.8431

2
2
2

0.0768
0.0821
0.0888

Likelihood Ratio
Score
Wald

Analysis of Maximum Likelihood Estimates

Parameter

DF

Estimate

Standard
Error

Chi-Square

Pr > ChiSq

Intercept
dummy1
dummy2

1
1
1

-1.0609
0.7732
1.1180

0.3867
0.5160
0.5138

7.5244
2.2454
4.7357

0.0061
0.1340
0.0295

Odds Ratio Estimates

Effect

Point
Estimate

dummy1
dummy2

2.167
3.059

95% Wald
Confidence Limits
0.788
1.117

5.957
8.373

Association of Predicted Probabilities and Observed Responses

Percent Concordant
Percent Discordant
Percent Tied
Pairs

46.0
22.2
31.7
2646

Somers' D
Gamma
Tau-a
c

0.238
0.349
0.115
0.619

Once again, lets look at the Odds Ratio estimates. Since dummy1 refers to Blacks, we
can conclude that Blacks are 2.167 times more likely to attempt suicide compared to
Whites. Since dummy2 refers to Hispanics, we can conclude that Hispanics are 3.059
times more likely to attempt suicide compared to Whites. Note that the 95%Confidence
Interval is significant only for Hispanics because it does not contain 1.000.